Brewster constraint series machine, system and medium for inverting surface water vapor impedance

Abstract

The present application relates to the technical field of surface water vapor impedance measurement method, especially to a surface water vapor impedance machine learning method using wave ratio constraint, a system and a medium. The method comprises: establishing a machine learning model of the regression error between the observed wave ratio and the actual, theoretical and virtual wave ratio, constructing a machine learning model between the observed aerodynamic impedance and the predicted aerodynamic impedance, creating a series machine between the observed surface water vapor impedance and the predicted surface water vapor impedance, and using the established series machine to invert the surface water vapor impedance in the unobserved data area. The present application is based on the physical link chain among the wave ratio, aerodynamic impedance and surface water vapor impedance, and constructs a progressive series machine for the inversion of the surface water vapor impedance in the unobserved data area. Compared with the traditional machine learning modeling method, the machine learning model established by the present application has higher training efficiency and more explicit physical meaning, and is a better combination of prior knowledge and artificial intelligence.

Description

Technical Field

[0001] This invention relates to the technical field of surface water vapor impedance measurement methods, and in particular to a machine learning method, system, and medium for surface water vapor impedance using Bowen ratio constraints. Background Technology

[0002] Surface water heat flux is the amount of water absorbed by terrestrial soils, vegetation, and water bodies from solar radiation that enters the atmosphere through soil evaporation, vegetation transpiration, vegetation intercepted evaporation, and water body evaporation. It includes physiological water consumption by vegetation, soil water excretion, surface water loss, and canopy interception losses. It forms the basis for applications in environmental planning, climate models, agricultural water conservation, drought monitoring, and water resource management. The surface water heat flux process is complex and is typically estimated using the Penman equation. The Penman equation requires numerous inputs related to heterogeneous surface temperature, surface saturated vapor pressure, atmospheric vapor pressure, surface aerodynamic impedance, and surface water vapor impedance.

[0003] In 1948, Penman integrated the energy and momentum driving force of water vapor through formula derivation, thus proposing the evaporation formula for free saturated water surfaces. Subsequently, Monteith introduced aerodynamic impedance and surface water vapor impedance using the concept of electrical resistance, providing a method for the evaporation of water from unsaturated water surfaces and land surfaces, known as the Penman-Montes formula. Although the Penman-Montes formula can be used to estimate the water and heat flux of ordinary underlying surfaces, and the physical meanings of the introduced aerodynamic impedance and surface water vapor impedance are clear, they cannot be derived into mathematical formulas due to the lack of a physical basis. To a certain extent, the complexity of the calculation is no less than that of the calculation of water and heat flux itself. As a result, to date, there are more than a dozen formulas for calculating aerodynamic impedance alone, and the estimation of surface water vapor impedance is even more difficult. Surface water vapor impedance is physically defined as the surface drag force when water escapes from the surface, and current estimations of surface water vapor impedance are mostly empirical, resulting in large errors.

[0004] Currently, observational data on surface water and heat fluxes are becoming increasingly abundant, covering different underlying surface types, various ecosystems, and multiple spatial scales. On the other hand, machine learning methods combined with artificial intelligence are becoming increasingly mature. To a certain extent, by acquiring new knowledge and reorganizing themselves, they continuously improve their performance and can uncover the implicit physical relationships between variables. The root cause of the large errors in current surface water vapor impedance estimation lies in the implicit nature of its relationships with other variables, which is precisely where machine learning excels. Therefore, using the abundant existing observational data on surface water and heat fluxes to train machine models to learn the implicit physical relationships between surface water vapor impedance and other variables provides a new approach for remote sensing inversion of surface water vapor impedance.

[0005] Therefore, this invention is proposed. Summary of the Invention

[0006] The purpose of this invention is to provide a Bowen ratio-constrained series machine, system, and medium for inverting surface water vapor impedance, in order to overcome or partially overcome the shortcomings of existing surface water vapor impedance estimation with large errors.

[0007] To overcome the above-mentioned technical problems, the technical solution adopted by the present invention is as follows:

[0008] In a first aspect, embodiments of the present invention provide a Bowen ratio-constrained series machine for inverting surface water vapor impedance, comprising the following steps:

[0009] S1: Establish a machine learning model for the regression error between observed Bowen ratio and actual, theoretical, and virtual Bowen ratios;

[0010] S2: Construct a machine learning model between observed aerodynamic impedance and predicted aerodynamic impedance;

[0011] S3: Create a series connection between the observed surface water vapor impedance and the predicted surface water vapor impedance;

[0012] S4: Use the established series machine to invert the surface water vapor impedance in areas without observation data.

[0013] In one embodiment, S1: establishing a machine learning model for the regression error between the observed Bowen ratio and the actual, theoretical, and virtual Bowen ratios includes the following steps:

[0014] S11: Establish a multiple regression model between the observed Bowen ratio and the actual, theoretical, and virtual Bowen ratios, specifically as follows:

[0015] β=[β a β t β v ][κ a κ t κ v ]'+b

[0016]

[0017]

[0018]

[0019]

[0020] In the formula, β represents the Bowen ratio, β a β represents the actual Bowen ratio. t β represents the theoretical Bowen ratio. v κ represents the virtual Bowen ratio, b represents the regression error of the multiple regression model, and κ represents the virtual Bowen ratio. a κ represents the regression coefficient of the actual Bowen ratio. tκ represents the regression coefficient of the theoretical Bowen ratio. v The regression coefficient representing the virtual Bowen ratio, Represent the actual BOV ratio, theoretical BOV ratio, and virtual BOV ratio of the i-th group obtained from remote sensing surface temperature data, respectively. si This represents the i-th surface temperature (K) obtained from remote sensing surface temperature data. and They represent T respectively si The minimum and maximum values ​​(K) of remotely sensed land surface temperature within the central window. and T represents respectively si The theoretical minimum and theoretical maximum surface temperatures (K) at that location. and T represents respectively si The virtual minimum and virtual maximum values ​​(K) of the surface temperature; This represents the Bowen ratio of the i-th observation obtained from the observation data. This represents the sensible heat flux observation data (W / m³). 2 ), ET i o This represents the observed surface water heat flux data (W / m³). 2 );

[0021] S12: Calculate the regression error between the observed Bowen ratio and the predicted Bowen ratio, specifically:

[0022]

[0023] In the formula, b i This indicates that the multiple regression model predicts... The regression error;

[0024] S13: Establish a machine learning model for the regression error between the observed Bowen ratio and the actual, theoretical, and virtual Bowen ratios, specifically as follows:

[0025] {b i} 1≤i≤I =M({v 1i ,…,v Ji} 1≤i≤I )

[0026] In the formula, v 1i , ..., v Ji Let J represent the J factors in the i-th group that affect the Bowen ratio prediction error, and M represent the established machine learning model used to learn the implicit physical relationship between the multivariate regression Bowen ratio prediction error and the J influencing factors.

[0027] In one embodiment, S2: constructing a machine learning model between observed aerodynamic impedance and predicted aerodynamic impedance includes the following steps:

[0028] S21: Obtain the observed aerodynamic impedance, specifically:

[0029]

[0030] In the formula, This represents the observed surface temperature (K). Indicates and The corresponding i-th observed air temperature (K), C a The volumetric heat capacity of air (J / (℃·m³)) 3 )), Indicates by and The inverted i-th observed aerodynamic impedance (s / m);

[0031] S22: The aerodynamic impedance is derived from S1, specifically as follows:

[0032]

[0033] In the formula, T represents the i-th predicted aerodynamic impedance (s / m) derived from S1. ai Indicates that it is used for exporting. The air temperature (K) R in the i-th reanalysis meteorological data ni Indicated by solar shortwave radiation, T si T ai The calculated net radiation (W / m) of the i-th generation 2 ), G i Represents the i-th soil heat flux (W / m³) 2 );

[0034] S23: Construct a machine learning model between observed aerodynamic impedance and predicted aerodynamic impedance, specifically:

[0035]

[0036] In the formula, m represents the established machine learning model used to learn the implicit physical relationship between observed aerodynamic impedance and predicted aerodynamic impedance.

[0037] In one embodiment, S3: creating a series connection between the observed surface water vapor impedance and the predicted surface water vapor impedance, specifically involves:

[0038] S31: Obtain the surface water vapor impedance, specifically:

[0039]

[0040] In the formula, This represents the calculated surface water vapor impedance (s / m) for the i-th observed surface. Indicates the use of The calculated slope (kPa / ℃) of the i-th observed air temperature-saturated vapor pressure is... Represents the soil heat flux (W / m³) of the i-th observation. 2 ), This represents the calculated observed saturated vapor pressure (kPa) of the i-th time. γ represents the actual water vapor pressure of the i-th air (kPa), and γ represents the wet / dry gauge constant (kPa / ℃).

[0041] S32: The predicted surface water vapor impedance is derived from S2, specifically as follows:

[0042]

[0043] In the formula, Δ represents the i-th predicted surface water vapor impedance (s / m) derived from S2. i Indicates the use of T ai The calculated slope (kPa / ℃) of the i-th actual air temperature-saturated vapor pressure is e. si e represents the i-th saturated vapor pressure (kPa) calculated using reanalysis meteorological data. ai This represents the actual air vapor pressure (kPa) of the i-th element in the reanalysis meteorological data. This represents the aerodynamic impedance (s / m) predicted using m;

[0044] S33: Create a series connection between the observed surface water vapor impedance and the predicted surface water vapor impedance, specifically:

[0045]

[0046] In the formula, Ψ represents the series connection between the observed surface water vapor impedance and the surface water vapor impedance derived from S2, which is used to learn the implicit physical relationship between the observed surface water vapor impedance and the predicted surface water vapor impedance.

[0047] In one embodiment, step S4: inverting the surface water vapor impedance of the area without observation data using the established series machine includes the following steps:

[0048] S41: Predict the Bowen ratio for areas without observed data, specifically:

[0049]

[0050]

[0051] In the formula, B represents the predicted Bowen ratio for the unobserved data area. These represent the actual BRVIN, theoretical BRVIN, virtual BRVIN, and T at pixel l, obtained from remotely sensed surface temperature data in areas without observational data. sl This represents the surface temperature (K) at pixel l obtained from remote sensing surface temperature data. and They represent T respectively sl The minimum and maximum values ​​(K) of remotely sensed land surface temperature within the central window. and T represents respectively sl The theoretical minimum and theoretical maximum surface temperatures (K) at that location. and T represents respectively sl The virtual minimum and virtual maximum values ​​of surface temperature (K), 1≤l≤L, where L represents the number of pixels in the area without observation data, v 1l , ..., v Jl This represents the J factors affecting the Bowen ratio at pixel l;

[0052] S42: Predict the aerodynamic impedance in areas without observational data, specifically:

[0053]

[0054] In the formula, T al R represents the air temperature (K) in the reanalysis meteorological data for areas without observational data. nl Indicated by solar shortwave radiation, T sl T al The calculated net radiation at pixel l (W / m) 2 ), G l Represents the soil heat flux at pixel l (W / m²) 2 );

[0055] S43: Inversion of surface water vapor impedance in areas with no observational data, specifically:

[0056]

[0057] In the formula, g sl This represents the surface water vapor impedance (s / m) for the inverted area without observational data.

[0058] In a second aspect, embodiments of the present invention provide a Bowen ratio-constrained series machine system for inverting surface water vapor impedance, used to execute the Bowen ratio-constrained series machine for inverting surface water vapor impedance as described in the first aspect above, comprising: a Bowen ratio module, an aerodynamic impedance module, a surface water vapor impedance module, and an inversion module;

[0059] The Bowen ratio module is used to execute the machine learning model described in the first aspect above for establishing the regression error between the observed Bowen ratio and the actual, theoretical, and virtual Bowen ratios;

[0060] The aerodynamic impedance module is used to perform the machine learning model described in the first aspect above, which establishes the relationship between observed aerodynamic impedance and predicted aerodynamic impedance.

[0061] The surface water vapor impedance module is used to perform the series connection machine described in the first aspect above, which creates the observed surface water vapor impedance and the predicted surface water vapor impedance.

[0062] The inversion module is used to perform the inversion of surface water vapor impedance in areas without observation data using a series machine as described in the first aspect above;

[0063] Based on the above technical solution, the Bowen ratio module further includes: a regression model unit, a regression error unit, and an error machine learning unit;

[0064] The regression model unit is used to execute the multivariate regression model established by the first aspect S11 above, which establishes the relationship between the observed Bowen ratio and the actual, theoretical, and virtual Bowen ratios.

[0065] The regression error unit is used to perform the calculation of the regression error between the observed Bovn ratio and the predicted Bovn ratio as described in the first aspect S12 above.

[0066] The error machine learning unit is used to execute the machine learning model described in the first aspect S13 above, which establishes the regression error between the observed Bowen ratio and the actual, theoretical, and virtual Bowen ratios.

[0067] Based on the above technical solution, the aerodynamic impedance module further includes: an aerodynamic impedance observation unit, an aerodynamic impedance prediction unit, and an aerodynamic impedance error learning unit.

[0068] The aerodynamic impedance observation unit is used to perform the acquisition of aerodynamic impedance as described in the first aspect S21 above.

[0069] The predicted aerodynamic impedance unit is used to perform the aerodynamic impedance derived from S1 as described in the first aspect S22 above.

[0070] The aerodynamic impedance error learning unit is used to perform the machine learning model described in the first aspect S23 above, which constructs a machine learning model between observed aerodynamic impedance and predicted aerodynamic impedance.

[0071] Based on the above technical solution, the surface water vapor impedance module further includes: a surface water vapor impedance observation unit, a surface water vapor impedance prediction unit, and a surface water vapor impedance error learning unit.

[0072] The surface water vapor impedance observation unit is used to perform the acquisition of surface water vapor impedance as described in the first aspect S31 above.

[0073] The predicted surface water vapor impedance unit is used to perform the above-mentioned first aspect S32, which derives the predicted surface water vapor impedance from S2.

[0074] The surface water vapor impedance error learning unit is used to perform the series machine for creating the observed surface water vapor impedance and the predicted surface water vapor impedance as described in the first aspect S33 above.

[0075] Based on the above technical solution, the inversion module further includes: a predictive Bowen ratio unit, a predictive aerodynamic impedance unit, and an inversion surface water vapor impedance unit.

[0076] The Bowen ratio prediction unit is used to perform the prediction of Bowen ratio for the unobserved data area as described in the first aspect S41 above.

[0077] The predicted aerodynamic impedance unit is used to perform the prediction of aerodynamic impedance in the region without observation data as described in the first aspect S42 above.

[0078] The surface water vapor impedance inversion unit is used to perform the inversion of surface water vapor impedance in the area without observation data as described in the first aspect S43 above.

[0079] Thirdly, embodiments of the present invention provide a computer-readable storage medium for inverting surface water vapor impedance by a Bowen ratio-constrained series machine, wherein a computer program is stored thereon, and when the computer program is executed by a processor, it implements the steps of the Bowen ratio-constrained series machine for inverting surface water vapor impedance as described in the first aspect.

[0080] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0081] 1) A method for inverting surface water vapor impedance by constraining machine learning using a physical model of surface water heat flux is proposed;

[0082] 2) By dividing the physical model into deterministic and non-deterministic parts, a machine model training approach is proposed that uses the deterministic part of the physical model to constrain the non-deterministic part of the physical model, thus breaking away from the previous pure data mining machine training mode.

[0083] 3) By splitting the physical model of surface water heat flux into three parts—Bowen ratio, aerodynamic impedance, and surface water vapor impedance—a progressive series engine was established that is linked to remote sensing parameters and meteorological variables.

[0084] 4) Based on the constraint of the physical link between Bowen ratio, aerodynamic impedance, and surface water vapor impedance, a series machine model for surface water vapor impedance inversion in areas without observation data was constructed.

[0085] 5) Compared with traditional machine learning modeling methods, the machine learning model with physical constraints established in this invention has higher training efficiency and clearer physical meaning, which is a better combination of human knowledge and artificial intelligence. Attached Figure Description

[0086] Figure 1 A schematic diagram of the process for a Bowen ratio-constrained series machine for inverting surface water vapor impedance provided in an embodiment of the present invention;

[0087] Figure 2 This is a flowchart illustrating the specific implementation of step 1 of the present invention;

[0088] Figure 3 This is a flowchart illustrating the specific implementation of step 2 of the present invention;

[0089] Figure 4 This is a flowchart illustrating the specific implementation of step 3 of the present invention;

[0090] Figure 5 This is a flowchart illustrating the specific implementation of step 4 of the present invention;

[0091] Figure 6 This is a schematic diagram of the structure of a Bowen ratio-constrained series machine system for inverting surface water vapor impedance provided in an embodiment of the present invention;

[0092] Figure 7 This is a schematic diagram of the structure of the Bowen ratio module described in this invention;

[0093] Figure 8 This is a schematic diagram of the aerodynamic impedance module described in this invention;

[0094] Figure 9 This is a schematic diagram of the surface water vapor impedance module described in this invention;

[0095] Figure 10 This is a schematic diagram of the inversion module described in this invention;

[0096] The attached diagram lists the components represented by each number as follows:

[0097] 1. Bowen Ratio Module; 2. Aerodynamic Impedance Module; 3. Surface Water Vapor Impedance Module; 4. Inversion Module; 11. Regression Model Unit; 12. Regression Error Unit; 13. Error Machine Learning Unit; 21. Observed Aerodynamic Impedance Unit; 22. Predicted Aerodynamic Impedance Unit; 23. Aerodynamic Impedance Error Learning Unit; 31. Observed Surface Water Vapor Impedance Unit; 32. Predicted Surface Water Vapor Impedance Unit; 33. Surface Water Vapor Impedance Error Learning Unit; 41. Predicted Bowen Ratio Unit; 42. Predicted Aerodynamic Impedance Unit; 43. Inverted Surface Water Vapor Impedance Unit. Detailed Implementation

[0098] In one embodiment, such as Figure 1 As shown, Figure 1 A schematic diagram of the process for a Bowen ratio-constrained series machine for inverting surface water vapor impedance provided in this embodiment of the invention includes the following steps:

[0099] S1: Establish a machine learning model for the regression error between observed Bowen ratio and actual, theoretical, and virtual Bowen ratios;

[0100] S2: Construct a machine learning model between observed aerodynamic impedance and predicted aerodynamic impedance;

[0101] S3: Create a series connection between the observed surface water vapor impedance and the predicted surface water vapor impedance;

[0102] S4: Use the established series machine to invert the surface water vapor impedance in areas without observation data;

[0103] The observed Bowen ratio, observed aerodynamic impedance, and observed surface water vapor impedance can be calculated from observation data of surface flux observation stations. These station data should cover different climate zones, different ecological types, different underlying surface characteristics, and different vegetation cover conditions. For example, the Haihe River Basin multi-scale surface flux and meteorological element observation dataset can be downloaded from the National Tibetan Plateau Scientific Data Center (https: / / data.tpdc.ac.cn / ). This dataset includes observation stations at Miyun, Daxing, Guantao, and Huailai stations; the underlying surface at Miyun station consists of orchards (plum and apple trees), corn / bare land, and urban areas. The data range covers January 1, 2008 to 201... On December 29, 2000, the underlying surface of Guantao Station was corn / wheat and cotton, with data covering the period from January 1, 2008 to December 31, 2010. The underlying surface of Daxing Station was corn / wheat and melons, with data covering the period from January 1, 2008 to December 31, 2010. The underlying surface of Huailai Station was corn, with data covering the period from January 1, 2021 to December 31, 2021. The observed variables included all variables required for establishing the machine model, such as air temperature, relative humidity, air pressure, wind speed, four-component radiation, surface radiation temperature, soil heat flux, water and heat flux, and sensible heat flux. The automatic weather station collected data at a frequency of 10 seconds, and the sensible heat flux and water and heat flux were observed every 30 minutes.

[0104] You can download the observation data and matrix for the alpine meadow ecosystem hydrothermal flux observation—Arou Superstation, Daman Superstation, Sidaoqiao Superstation, and the alpine meadow-grassland hybrid superstation of the Qinghai Lake Basin Surface Processes Integrated Observation Network—from the National Qinghai-Tibet Plateau Scientific Data Center (https: / / data.tpdc.ac.cn / ). Specifically, the Arou Superstation observation data covers January 1, 2015 to December 31, 2017, with an underlying alpine grassland; the Daman Superstation observation data covers January 1, 2021 to December 31, 2021, with an underlying maize surface; and the Sidaoqiao Superstation observation data covers January 1, 2015 to December 31, 2021. From September 3, 2018 to December 31, 2019, the underlying surface was tamarisk. The mixed superstation data covered the period from September 3, 2018 to December 31, 2019, where the underlying surface was a mixture of alpine meadow and alpine grassland. The observation matrix covered the period from May 10, 2012 to September 26, 2012, where the underlying surface was cornfield. The observed variables included all variables required for building the machine model, such as air temperature and humidity, air pressure, wind speed, surface radiation temperature, soil heat flux, four-component radiation, sensible heat flux, and hydrothermal flux. The observation frequency of meteorological elements was to ensure 144 data points per day (every 10 minutes), and the flux observation frequency was once every 30 minutes of logger output.

[0105] The soil moisture-atmosphere coupled experiment (SMACEX02) observation dataset can be downloaded from the global flux network (https: / / fluxnet.fluxdata.org / about). The data covers the period from June 19 to July 9, 2002. The core area of ​​the experiment is the Walnut Creek watershed, with maize and soybean as the main crops. Twelve eddy covariance observation devices were set up, with six set up in the maize field and six set up in the soybean field. The observed variables include wind speed, wind direction, air temperature, water vapor pressure, soil temperature, humidity, sensible heat flux, water and heat flux, and soil heat flux. The data frequency is once every 30 minutes.

[0106] Remote sensing imagery can be selected from the Landsat series. The Landsat series began in August 1972, and Landsat 8 and 9 are still in orbit, with an observation period of 50 years. The various Landsat sensors are very similar in terms of wavelength, bandwidth, number of bands, temporal resolution, and spatial resolution, and their transit times at the same local time are very close, ensuring the consistency of Landsat Earth observations over time. All data is open source and can be downloaded for free from https: / / earthdata.nasa.gov / . For the inversion of various surface parameters from Landsat satellite images, relevant literature can be consulted, such as surface temperature (Mhawej and Ghaleb, 2020), surface albedo (Liang et al., 1999), surface emissivity (Zhang et al., 2017), vegetation index, and algorithms for vegetation cover (Liu Zhenwei, 2021).

[0107] References

[0108] Liang SL, Strahler AH and Walthall C., Retrieval of Land SurfaceAlbedo from Satellite Observations: A Simulation Study, Journal of AppliedMeteorology 38(6):712-725, 1999.

[0109] Mario Mhawej, Ghaleb Faour, Open-source Google Earth Engine 30-mevapotranspiration rates retrieval: The SEBALIGEE system, Environmental Modeling and Software 133, 104845, 2020.

[0110] Liu Zhenwei, Research on the Construction of Long-Term High-Resolution Soil Moisture Dataset for the Qinghai-Tibet Plateau Based on GEE, Master's Thesis, University of Chinese Academy of Sciences, 2021;

[0111] Zhang Xin, Ding Feng, Peng Xianli, Wu Wenfeng, Fan Pengyu, Surface emissivity inversion and IDL implementation based on Landsat data, 55-55, 2017.

[0112] Based on the above embodiments, in some embodiments of the present invention, further, such as Figure 2 As shown, S1: Establishing a machine learning model for the regression error between the observed Bowen ratio and the actual, theoretical, and virtual Bowen ratios. One implementation method includes the following steps:

[0113] S11: Establish a multiple regression model between the observed Bowen ratio and the actual, theoretical, and virtual Bowen ratios, specifically as follows:

[0114] β=[β a β t β v ][κ a κ t κ v ]'+b

[0115]

[0116]

[0117]

[0118]

[0119] In the formula, β represents the Bowen ratio, β a β represents the actual Bowen ratio. t β represents the theoretical Bowen ratio. v κ represents the virtual Bowen ratio, b represents the regression error of the multiple regression model, and κ represents the virtual Bowen ratio. a κ represents the regression coefficient of the actual Bowen ratio. t κ represents the regression coefficient of the theoretical Bowen ratio. v The regression coefficient representing the virtual Bowen ratio, Represent the actual BOV ratio, theoretical BOV ratio, and virtual BOV ratio of the i-th group obtained from remote sensing surface temperature data, respectively. si This represents the i-th surface temperature (K) obtained from remote sensing surface temperature data. and They represent T respectively si The minimum and maximum values ​​(K) of remotely sensed land surface temperature within the central window. and T represents respectivelysi The theoretical minimum and theoretical maximum surface temperatures (K) at that location. and T represents respectively si The virtual minimum and virtual maximum values ​​(K) of the surface temperature; This represents the Bowen ratio of the i-th observation obtained from the observation data. This represents the sensible heat flux observation data (W / m³). 2 ), ET i o This represents the observed surface water heat flux data (W / m³). 2 );

[0120] Specifically, since the Landsat return cycle is 16 days, the parameters of Landsat remote sensing surface temperature, surface albedo, vegetation index, and vegetation cover corresponding to the observation locations and observation times of each station can be extracted through the Google Earth online platform. For specific operations, please refer to the literature Mhawej and Ghaleb (2020). The temporal number of each remote sensing parameter extracted is I.

[0121] Actual Bowen ratio The calculation can be referenced (Long and Singh, 2012), and the theoretical Bowen ratio is calculated accordingly. The virtual BOV ratio can be calculated using reference (Zhang et al., 2008), and the virtual BOV ratio can be calculated using reference (Yang Yongmin et al., 2016).

[0122] Long,D.,and Singh,VP,A Two-source Trapezoid Model for Evapotranspiration(TTME) from satellite imagery.Remote Sensing of Environment,2012,121,370–388.

[0123] Renhua Zhang, Jing Tian, ​​Hongbo Su, Xiaomin Sun, Shaohui Chen and JunXia, Two Improvements of an Operational Two-Layer Model for TerrestrialSurface Heat Flux Retrieval, Sensors 2008, 8, 6165-6187; DOI: 10.3390 / s8106165.

[0124] Yang Yongmin, Li Lu, Pang Zhiguo, et al. Construction and verification of remote sensing evapotranspiration model based on theoretical parameter space. Remote Sensing Technology and Application, 2016, 31(2), 324-331.

[0125] S12: Calculate the regression error between the observed Bowen ratio and the predicted Bowen ratio, specifically:

[0126]

[0127] In the formula, b i This indicates that the multiple regression model predicts... The regression error;

[0128] S13: Establish a machine learning model for the regression error between the observed Bowen ratio and the actual, theoretical, and virtual Bowen ratios, specifically as follows:

[0129] {b i} 1≤i≤I =M({v 1i ,…,v Ji} 1≤i≤I )

[0130] In the formula, v 1i , ..., v Ji Let J represent the J factors in the i-th group that affect the Bowen ratio prediction error, and M represent the established machine learning model used to learn the implicit physical relationship between the multivariate regression Bowen ratio prediction error and the J influencing factors.

[0131] Specifically, the machine learning model can be selected from one of random forest, convolutional network, perceptron neural network or deep learning model.

[0132] Based on the above embodiments, in some embodiments of the present invention, further, such as Figure 3 As shown, S2: Constructing a machine learning model between observed aerodynamic impedance and predicted aerodynamic impedance, one implementation method includes the following steps:

[0133] S21: Obtain the observed aerodynamic impedance, specifically:

[0134]

[0135] In the formula, This represents the observed surface temperature (K). Indicates and The corresponding i-th observed air temperature (K), C a The volumetric heat capacity of air (J / (℃·m³)) 3 )), Indicates by and The inverted i-th observed aerodynamic impedance (s / m);

[0136] S22: The aerodynamic impedance is derived from S1, specifically as follows:

[0137]

[0138] In the formula, T represents the i-th predicted aerodynamic impedance (s / m) derived from S1. ai Indicates that it is used for exporting. The air temperature (K) R in the i-th reanalysis meteorological data ni Indicated by solar shortwave radiation, T si T ai The calculated net radiation (W / m) of the i-th generation 2 ), G i Represents the i-th soil heat flux (W / m³) 2 );

[0139] Specifically, the air temperature in the reanalysis meteorological data can be obtained from a high-resolution meteorological driving dataset for the Chinese region, covering the period from 1979 to 2018, with a spatial resolution of 0.1° and a temporal resolution of 3 hours, which is currently the highest spatial resolution.

[0140] For calculation of soil heat flux, please refer to the reference (Zhu Weiwei, 2014);

[0141] Zhu Weiwei, Research on Remote Sensing Estimation Methods for Soil Heat Flux, Doctoral Dissertation, University of Chinese Academy of Sciences, p. 47, 2014.

[0142] S23: Construct a machine learning model between observed aerodynamic impedance and predicted aerodynamic impedance, specifically:

[0143]

[0144] In the formula, m represents the established machine learning model used to learn the implicit physical relationship between observed aerodynamic impedance and predicted aerodynamic impedance;

[0145] Specifically, the machine learning model may be the same as or different from other machine models, and those skilled in the art can make the choice based on the specific circumstances.

[0146] Based on the above embodiments, in some embodiments of the present invention, further, such as Figure 4 As shown, S3: creating a series connection between the observed surface water vapor impedance and the predicted surface water vapor impedance, one implementation method includes the following steps:

[0147] S31: Obtain the observed surface water vapor impedance, specifically:

[0148]

[0149] In the formula, This represents the calculated surface water vapor impedance (s / m) for the i-th observed surface. Indicates the use of The calculated slope (kPa / ℃) of the i-th observed air temperature-saturated vapor pressure is... Represents the soil heat flux (W / m³) of the i-th observation. 2 ), This represents the calculated observed saturated vapor pressure (kPa) of the i-th time. The γ represents the actual water vapor pressure of the i-th air element (kPa), and the γ represents the tween constant (kPa / ℃). The saturated water vapor pressure... Calculations can be made using reference (Jonh et al., 2022);

[0150] Jonh Billy Silva, Denilton Carlos Gaio, et al., Evaluation of methods for estimating atmospheric emissivity in Mato-Grossense Cerrado, https: / / www.redalyc.org / journal / 928 / 92860487003 / html / , p4, 2022.

[0151] S32: The predicted surface water vapor impedance is derived from S2, specifically as follows:

[0152]

[0153] In the formula, Δ represents the i-th predicted surface water vapor impedance (s / m) derived from S2. i Indicates the use of T ai The calculated slope (kPa / ℃) of the i-th actual air temperature-saturated vapor pressure is e. si e represents the i-th saturated vapor pressure (kPa) calculated using reanalysis meteorological data. ai This represents the actual air vapor pressure (kPa) of the i-th element in the reanalysis meteorological data. This represents the aerodynamic impedance (s / m) predicted using m;

[0154] S33: Create a series connection between the observed surface water vapor impedance and the predicted surface water vapor impedance, specifically:

[0155]

[0156] In the formula, Ψ represents the series connection between the observed surface water vapor impedance and the surface water vapor impedance derived from S2, which is used to learn the implicit physical relationship between the observed surface water vapor impedance and the predicted surface water vapor impedance.

[0157] Based on the above embodiments, in some embodiments of the present invention, further, such as Figure 5 As shown, S4: using the established series machine to invert the surface water vapor impedance in areas without observation data, one implementation method includes the following steps:

[0158] S41: Predict the Bowen ratio for areas without observed data, specifically:

[0159]

[0160]

[0161] In the formula, B represents the predicted Bowen ratio for the unobserved data area. These represent the actual BRVIN, theoretical BRVIN, virtual BRVIN, and T at pixel l, obtained from remotely sensed surface temperature data in areas without observational data. sl This represents the surface temperature (K) at pixel l obtained from remote sensing surface temperature data. and They represent T respectively sl The minimum and maximum values ​​(K) of remotely sensed land surface temperature within the central window. and T represents respectively sl The theoretical minimum and theoretical maximum surface temperatures (K) at that location. and T represents respectively sl The virtual minimum and virtual maximum values ​​of surface temperature (K), 1≤l≤L, where L represents the number of pixels in the area without observation data, v 1l , ..., v Jl This represents the J factors affecting the Bowen ratio at pixel l;

[0162] The area without observation data refers to the area where there are no observations of water and heat flux, sensible heat flux, etc., but surface temperature, albedo, vegetation cover, and surface emissivity can be obtained from remote sensing data.

[0163] S42: Predict the aerodynamic impedance in areas without observational data, specifically:

[0164]

[0165] In the formula, T al R represents the air temperature (K) in the reanalysis meteorological data for areas without observational data. nlIndicated by solar shortwave radiation, T sl T al The calculated net radiation at pixel l (W / m) 2 ), G l Represents the soil heat flux at pixel l (W / m²) 2 );

[0166] S43: Inversion of surface water vapor impedance in areas with no observational data, specifically:

[0167]

[0168] In the formula, g sl This represents the surface water vapor impedance (s / m) for the inverted area without observational data.

[0169] In summary, it is possible to invert surface water vapor impedance in areas without observational data.

[0170] In one embodiment, such as Figure 6 As shown, a Bowen ratio-constrained series machine system for inverting surface water vapor impedance is provided, which is used to execute the Bowen ratio-constrained series machine for inverting surface water vapor impedance as described in the above embodiment, including: Bowen ratio module 1, aerodynamic impedance module 2, surface water vapor impedance module 3, and inversion module 4.

[0171] The Bowen ratio module 1 is used to execute the machine learning model described in the above embodiments for establishing the regression error between the observed Bowen ratio and the actual, theoretical, and virtual Bowen ratios;

[0172] The aerodynamic impedance module 2 is used to execute the machine learning model described in the above embodiment for constructing the relationship between observed aerodynamic impedance and predicted aerodynamic impedance.

[0173] The surface water vapor impedance module 3 is used to perform the series connection machine described in the above embodiment for creating the observed surface water vapor impedance and the predicted surface water vapor impedance.

[0174] The inversion module 4 is used to perform the inversion of surface water vapor impedance in areas without observation data using a series machine as described in the above embodiment;

[0175] In one embodiment, such as Figure 7 As shown, the Bowen ratio module 1 includes: a regression model unit 11, a regression error unit 12, and an error machine learning unit 13;

[0176] The regression model unit 11 is used to execute the multivariate regression model established in S11 of the above embodiment to establish the relationship between the observed Bowen ratio and the actual, theoretical and virtual Bowen ratios.

[0177] The regression error unit 12 is used to perform the calculation of the regression error between the observed Bovn ratio and the predicted Bovn ratio as described in S12 of the above embodiment;

[0178] The error machine learning unit 13 is used to execute the machine learning model described in S13 of the above embodiment to establish the regression error between the observed Bowen ratio and the actual, theoretical and virtual Bowen ratios;

[0179] In one embodiment, such as Figure 8 As shown, the aerodynamic impedance module 2 includes: an aerodynamic impedance observation unit 21, an aerodynamic impedance prediction unit 22, and an aerodynamic impedance error learning unit 23.

[0180] The aerodynamic impedance observation unit 21 is used to perform the acquisition of aerodynamic impedance as described in S21 of the above embodiment;

[0181] The predicted aerodynamic impedance unit 22 is used to perform the aerodynamic impedance derived from S1 as described in S22 of the above embodiment;

[0182] The aerodynamic impedance error learning unit 23 is used to execute the machine learning model constructed between the observed aerodynamic impedance and the predicted aerodynamic impedance as described in S23 of the above embodiment;

[0183] In one embodiment, such as Figure 9 As shown, the surface water vapor impedance module 3 includes: a surface water vapor impedance observation unit 31, a surface water vapor impedance prediction unit 32, and a surface water vapor impedance error learning unit 33.

[0184] The surface water vapor impedance observation unit 31 is used to perform the acquisition of surface water vapor impedance as described in S31 of the above embodiment.

[0185] The predicted surface water vapor impedance unit 32 is used to perform the derived predicted surface water vapor impedance from S2 as described in S32 of the above embodiment.

[0186] The surface water vapor impedance error learning unit 33 is used to execute the series machine for creating the observed surface water vapor impedance and the predicted surface water vapor impedance as described in S33 of the above embodiment.

[0187] In one embodiment, such as Figure 10 As shown, the inversion module 4 includes: a predictive Bowen ratio unit 41, a predictive aerodynamic impedance unit 42, and an inversion surface water vapor impedance unit 43.

[0188] The Bowen ratio prediction unit 41 is used to perform the prediction of Bowen ratio for the unobserved data area as described in S41 of the above embodiment.

[0189] The predicted aerodynamic impedance unit 42 is used to perform the prediction of aerodynamic impedance in the region without observation data as described in S42 of the above embodiment.

[0190] The surface water vapor impedance inversion unit 43 is used to perform the inversion of surface water vapor impedance in the area without observation data as described in S43 of the above embodiment.

[0191] This invention also provides a computer-readable storage medium for a Bowen ratio-constrained series machine for inverting surface water vapor impedance. The medium stores a computer program that, when executed by a processor, can implement the Bowen ratio-constrained series machine for inverting surface water vapor impedance provided in this invention. For example, when the computer program is executed by the processor, it implements... Figures 1-5 The technical solutions of any of the illustrated series machine embodiments are similar in principle and technical effect, and will not be described again here.

[0192] The advantages of the Bowen ratio-constrained series machine for inverting surface water vapor impedance proposed in this invention are as follows:

[0193] 1) The physical model constrains the machine learning training, making the machine learning model more focused on learning the relationships between variables that do not have explicit physical equations;

[0194] 2) Constraints based on deterministic relationships in the physical model can reduce the complexity of relationships between machine learning variables;

[0195] 3) A series model of a machine with progressively advancing Bowen ratio, aerodynamic impedance, and surface water vapor impedance was established;

[0196] 4) Combining existing practical, theoretical, and virtual Bowenbi knowledge with machine learning is more efficient than traditional machine learning training methods;

[0197] 5) It can use conventional surface parameter remote sensing products and meteorological reanalysis data to realize the surface water vapor impedance in areas without surface water and heat flux observation, and then estimate the sensible heat flux and water and heat flux.

[0198] The embodiment shown in the Bowen ratio-constrained series machine system for inverting surface water vapor impedance and the computer storage medium is the same as the series machine embodiment; please refer to the series machine embodiment. Those skilled in the art can make equivalent substitutions and modifications to this invention without creative effort, but these are also within the protection scope of this invention.

Claims

1. A method for inverting surface water vapor impedance using a Bowen ratio-constrained series machine, characterized in that, Includes the following steps: S1: Establish a machine learning model for the regression error between the observed Bowen ratio and the actual, theoretical, and virtual Bowen ratios, including the following steps: S11: Establish a multiple regression model between the observed Bowen ratio and the actual, theoretical, and virtual Bowen ratios, specifically as follows: In the formula, β represents the Bowen ratio, β a β represents the actual Bowen ratio. t β represents the theoretical Bowen ratio. v κ represents the virtual Bowen ratio, b represents the regression error of the multiple regression model, and κ represents the virtual Bowen ratio. a κ represents the regression coefficient of the actual Bowen ratio. t κ represents the regression coefficient of the theoretical Bowen ratio. v The regression coefficient representing the virtual Bowen ratio, , , Represent the actual BOV ratio, theoretical BOV ratio, and virtual BOV ratio of the i-th group obtained from remote sensing surface temperature data, respectively. si This represents the i-th surface temperature (K) obtained from remote sensing surface temperature data. and They represent T respectively si The minimum and maximum values ​​(K) of remotely sensed surface temperature within the central window. and T represents respectively si The theoretical minimum and theoretical maximum surface temperature (K). and T represents respectively si The virtual minimum and virtual maximum values ​​of surface temperature (K); This represents the Bowen ratio of the i-th observation obtained from the observation data. This represents the sensible heat flux observation data (W / m³). 2 ), This represents the i-th surface water heat flux observation data (W / m³). 2 ); S12: Calculate the regression error between the observed Bowen ratio and the predicted Bowen ratio, specifically: 1≤i≤I In the formula, b i This indicates that the multiple regression model predicts... The regression error; S13: Establish a machine learning model for the regression error between the observed Bowen ratio and the actual, theoretical, and virtual Bowen ratios, specifically as follows: In the formula, v 1i ,···,v Ji Let J represent the J factors in the i-th group that affect the Bowen ratio prediction error, and M represent the established machine learning model used to learn the implicit physical relationship between the multivariate regression Bowen ratio prediction error and the J influencing factors. S2: Construct a machine learning model between observed aerodynamic impedance and predicted aerodynamic impedance; S3: Create a series connection between the observed surface water vapor impedance and the predicted surface water vapor impedance; S4: Use the established series machine to invert the surface water vapor impedance in areas without observation data.

2. The method according to claim 1, characterized in that, S2 includes the following steps: S21: Obtain the observed aerodynamic impedance, specifically: 1≤i≤I In the formula, This represents the observed surface temperature (K). Indicates and The corresponding i-th observed air temperature (K), C a The volumetric heat capacity of air (J / (℃·m)) 3 )), Indicates by , and The inverted i-th observed aerodynamic impedance (s / m); S22: The aerodynamic impedance is derived from S1, specifically as follows: In the formula, T represents the i-th predicted aerodynamic impedance (s / m) derived from S1. ai Indicates that it is used for exporting. The air temperature (K) R in the i-th reanalysis meteorological data ni Indicated by solar shortwave radiation, T si T ai The calculated net radiation (W / m) of the i-th generation 2 ), G i Represents the i-th soil heat flux (W / m³) 2 ); S23: Construct a machine learning model between observed aerodynamic impedance and predicted aerodynamic impedance, specifically: In the formula, m represents the established machine learning model used to learn the implicit physical relationship between observed aerodynamic impedance and predicted aerodynamic impedance.

3. The method according to claim 1, characterized in that, S3 includes the following steps: S31: Obtain the observed surface water vapor impedance, specifically: In the formula, This represents the calculated surface water vapor impedance (s / m) for the i-th observed surface. Indicates the use of The calculated slope (kPa / ℃) of the i-th observed air temperature-saturated vapor pressure is as follows: Represents the soil heat flux (W / m³) of the i-th observation. 2 ), This represents the calculated observed saturated vapor pressure (kPa) of the i-th time. γ represents the actual water vapor pressure of the i-th air (kPa), and γ represents the wet / dry gauge constant (kPa / °C). S32: The predicted surface water vapor impedance is derived from S2, specifically as follows: In the formula, This represents the i-th predicted surface water vapor impedance (s / m) derived from S2. Indicates the use of T ai The calculated slope (kPa / ℃) of the i-th actual air temperature-saturated vapor pressure is e. si e represents the i-th saturated vapor pressure (kPa) calculated using reanalysis meteorological data. ai This represents the actual air vapor pressure (kPa) of the i-th element in the reanalysis meteorological data. This represents the aerodynamic impedance (s / m) predicted using m. S33: Create a series connection between the observed surface water vapor impedance and the predicted surface water vapor impedance, specifically: In the formula, Ψ represents the series connection between the observed surface water vapor impedance and the surface water vapor impedance derived from S2, which is used to learn the implicit physical relationship between the observed surface water vapor impedance and the predicted surface water vapor impedance.

4. The method according to claim 1, characterized in that, S4 includes the following steps: S41: Predict the Bowen ratio for areas without observed data, specifically: In the formula, B represents the predicted Bowen ratio for the unobserved data area. , , These represent the actual BRVIN, theoretical BRVIN, and virtual BRVIN at pixel l, obtained from remotely sensed surface temperature data in areas without observation data. sl This represents the surface temperature (K) at pixel l obtained from remote sensing surface temperature data. and They represent T respectively sl The minimum and maximum values ​​(K) of remotely sensed surface temperature within the central window. and T represents respectively sl The theoretical minimum and theoretical maximum surface temperature (K). and T represents respectively sl The virtual minimum and virtual maximum values ​​of surface temperature (K), 1≤l≤L, where L represents the number of pixels in the area without observation data, v 1l ,···,v Jl This represents the J factors affecting the Bowen ratio at pixel l; S42: Predict the aerodynamic impedance in areas without observational data, specifically: In the formula, T al R represents the air temperature (K) in the reanalysis meteorological data for areas without observational data. nl Indicated by solar shortwave radiation, T sl T al The calculated net radiation at pixel l (W / m) 2 ), G l Represents the soil heat flux at pixel l (W / m²) 2 ); S43: Inversion of surface water vapor impedance in areas with no observational data, specifically: In the formula, g sl This represents the surface water vapor impedance (s / m) for the inverted area without observational data.

5. A Bowen ratio-constrained series machine system for inverting surface water vapor impedance, characterized in that, A Bowen ratio-constrained series machine method for performing the inversion of surface water vapor impedance according to any one of claims 1-4, comprising: a Bowen ratio module, an aerodynamic impedance module, a surface water vapor impedance module, and an inversion module; The Bowen ratio module is used to execute the machine learning model that establishes the regression error between the observed Bowen ratio and the actual, theoretical, and virtual Bowen ratios. The aerodynamic impedance module is used to execute the machine learning model that builds the relationship between observed aerodynamic impedance and predicted aerodynamic impedance. The surface water vapor impedance module is used to execute the series connection mechanism for creating the observed surface water vapor impedance and the predicted surface water vapor impedance. The inversion module is used to perform the inversion of surface water vapor impedance in areas without observation data using a series machine; The Bowen ratio module includes: a regression model unit, a regression error unit, and an error machine learning unit; The regression model unit is used to execute the multivariate regression model described in S11, which establishes the relationship between the observed Bowen ratio and the actual, theoretical, and virtual Bowen ratios. The regression error unit is used to perform the calculation of the regression error between the observed Bowen ratio and the predicted Bowen ratio as described in S12. The error machine learning unit is used to execute the machine learning model described in S13 for establishing the regression error between the observed Bowen ratio and the actual, theoretical, and virtual Bowen ratios. The aerodynamic impedance module includes: an aerodynamic impedance observation unit, an aerodynamic impedance prediction unit, and an aerodynamic impedance error learning unit. The aerodynamic impedance observation unit is used to perform the acquisition of aerodynamic impedance as described in S21. The predicted aerodynamic impedance unit is used to perform the aerodynamic impedance derived from S1 as described in S22. The aerodynamic impedance error learning unit is used to perform the machine learning model described in S23, which builds a machine learning model between the observed aerodynamic impedance and the predicted aerodynamic impedance. The surface water vapor impedance module includes: a surface water vapor impedance observation unit, a surface water vapor impedance prediction unit, and a surface water vapor impedance error learning unit. The surface water vapor impedance observation unit is used to perform the acquisition of surface water vapor impedance as described in S31. The predicted surface water vapor impedance unit is used to perform the process described in S32, which derives the predicted surface water vapor impedance from S2. The surface water vapor impedance error learning unit is used to execute the series machine described in S33 for creating the observed surface water vapor impedance and the predicted surface water vapor impedance. The inversion module includes: a predictive Bowen ratio unit, a predictive aerodynamic impedance unit, and an inversion surface water vapor impedance unit. The Bowen ratio prediction unit is used to perform the prediction of the Bowen ratio of the unobserved data area as described in S41. A predictive aerodynamic impedance unit is used to perform the prediction of aerodynamic impedance in the region without observation data as described in S42. The surface water vapor impedance inversion unit is used to perform the inversion of surface water vapor impedance in the area without observation data as described in S43.

6. A Bovn ratio-constrained series machine computer-readable storage medium for inverting surface water vapor impedance, wherein a computer program is stored thereon, characterized in that, When the computer program is executed by the processor, it implements the method of inverting the Bowen ratio-constrained series machine for surface water vapor impedance as described in any one of claims 1-4.

Images